Goodness-of-fit tests for the weibull distribution with unknown parameters and heavy censoring

Abstract

Goodness-of-fit tests are considered for testing the two-parameters Weibull distribution based on type II censored sampling with both parameters assumed unknown. Some extremely heavy censoring levels are considered which are useful when analyzing in-service field data with a large population and a small number of failures. Critical values are obtained by Monte Carlo simulation for Kolmogorov-Smirnov, Kuiper and Cramer-von Mises type test statistics. The approximate Snedecor’s F -distribution is verified for the Mann-Scheuer-Fertig test statistic. The two-sided Mann-Scheuer-Fertig test is also studied and found to be important. A power study is carried out for these four test statistics for both moderate and heavy censoring for a number of different alternative models.